Testing
Serum Lp(a) predicts progression-free survival and severe adverse events in 227 lung adenocarcinoma patients on first-line chemoimmunotherapy (Front Immunol 2026)
Original title: Development of a lipoprotein(a)-based model for predicting progression-free survival and grade3/4 adverse events in driver gene negative metastatic lung adenocarcinoma patients with PD-L1 TPS <50
This prospective cohort study evaluated lipoprotein(a) as a biomarker in 227 driver-gene-negative metastatic lung adenocarcinoma patients with PD-L1 tumour proportion score below 50%, receiving first-line chemoimmunotherapy, split 7:3 into training and internal validation sets. A LASSO and multivariate Cox model built on serum LPA predicted 365-day progression-free survival with an AUC of 0.78 (0.62-0.94) in training and 0.95 (0.84-1.00) in validation, with higher LPA independently predicting disease progression. A separate machine-learning model, with AdaBoost outperforming random forest, elastic-net, LASSO and support vector machine on accuracy, precision, recall and F1, predicted grade 3/4 adverse events from the same biomarker panel. This is an oncology application outside Lp(a)'s usual cardiovascular context, proposing it as a prognostic and toxicity biomarker for chemoimmunotherapy response; the cohort is modest and the model needs external validation.
Original abstract
Objective: This study evaluated the value of lipoprotein(a) (LPA) in lung adenocarcinoma (LUAD) patients receiving first-line chemoimmunotherapy and developed a model to predict progression-free survival (PFS) and grade 3/4 adverse events (G3/4 AEs).
Methods: A prospective cohort study was conducted on driver gene negative metastatic LUAD patients with PD-L1 TPS <50%, who received first-line chemoimmunotherapy. The data were randomly sampled into training and internal validation sets following a 7:3 proportion. We constructed a prognostic model for progression-free survival (PFS) via LASSO and multivariate Cox regression analyses. We explored five methods-random forest, AdaBoost, elastic-net, LASSO, and support vector machine (SVM)-to develop a prediction model for G3/4 AEs.
Results: A total of 227 patients completed the follow-up. The AUC was 0.78(0.62-0.94) for 365-day PFS in the training cohort and 0.95(0.84-1.00) in the internal validation cohort. The serum LPA level independently predicted disease progression in patients receiving first-line chemoimmunotherapy. AdaBoost outperformed other machine learning methods in terms of accuracy, precision, recall, and F1 scores on both the training and validation sets, leading to its selection for the final G3/4 AE prediction model.
Conclusion: High LPA expression in the serum was a risk factor for metastatic driver gene-negative lung adenocarcinoma patients receiving first-line chemoimmunotherapy. Our models had favorable value in predicting PFS and G3/4 AEs, which might assist in identifying patients less likely to benefit from initial chemoimmunotherapy.
Summary written by lp-a.org from the published abstract; figures as published. Page updated 17 August 2026. Methods.